The present invention relates to a natural language dialogue apparatus and method to effectively output a response for a question to ask a value of truth or falsehood using natural language from a user.
Recently, a natural language interface to accept a natural language input such as speech or text is realized. An expert system using this natural language interface is also developed. Except for the expert system, an apparatus to accept input such as speech or text is utilized as an interface equal to a level of conversation between humans. As an example of this interface, a car navigation system is given. In the car navigation system, a user can operate by his voice during driving. In short, the car navigation system is used as an speech input system of natural language. In case of accepting the natural language input such as speech or text, a computer side must receive various kinds of question formats. As a format of high use frequency in the various kinds of question formats, a question for value of truth or falsehood for the user to confirm some proposition is given. In the question for value of truth or falsehood, the user presents the proposition to be confirmed and asks the truth or falsehood. Content of the proposition is the user""s belief or action. Concretely, in case of car navigation task, the following question is given as an example. xe2x80x9cDo I drive in this lane?xe2x80x9d
In this question, during driving on a road including a plurality of lanes, the user has the intention to confirm whether the present lane is suitable for the next movement such as a right turn or a left turn to arrive at the destination. If the user previously confirms a suitable lane, he need not hurriedly change lanes just before the next movement. As a result, the user is safety during driving increases. Furthermore, the user can drive in safety.
As for the above-mentioned question, a response process of a system for the value of truth or falsehood is explained. FIG. 1 is a flow chart of processing of the system to respond to the value of truth or falsehood according to the prior art.
(1) First, the present lane of car""s driving is confirmed in order to recognize xe2x80x9cthis lanexe2x80x9d. In short, the present status of the car""s driving is recognized (S11).
(2) According to previously known data of the present place, the set course, and the destination, a lane suitable for the next movement is calculated. In short, an answer for the question is calculated (S12).
(3) It is decided whether the present lane recognized at S11 coincides with the suitable lane calculated at S12. According to this decision, a value of truth or falsehood is obtained (S13).
(4) The value of truth or falsehood obtained at S13, and the answer obtained at S12 if necessary, are outputted as response data (S14).
However, above-mentioned prior art includes the following problems. In short, in the prior art, it is difficult to respond to the question for the value of truth or falsehood. The reason is explained as follows.
First, a processing load to confirm the lane in above process (1) is high. In general, in the car navigation system, GPS (Global Positioning System) is used to recognize a position of the car. In accuracy of present GPS, errors such as several meters or several tens of meters occur. Therefore, it is impossible to recognize the lane by GPS. As a result, the present car navigation system can not respond to the above-mentioned question for value of truth or falsehood. In order to confirm this lane, a special sensor is prepared in addition to the components of the car navigation system. However, even if a dialogue apparatus to cope with the question for value of truth or falsehood is realized, the component of the apparatus is expensive. On the other hand, if the natural language dialogue system does not accept the question of value of truth or falsehood, the user can not speak this kind of question at a timing to be confirmed. Accordingly, the quality of the dialogue interface goes down.
Second, in addition to output by retrieving the answer (solution) at S12, a calculation processing to decide whether the answer at S13 corresponds to a value of truth or falsehood of the question must be executed. Therefore, generation processing of response sentence is complicated.
This kind of problem occurs without limitation to the car navigation system. For example, in a load guidance system set on a service area, assume that the user speaks the question of value of truth or falsehood as xe2x80x9cDo I drive along this direction to go to the destination?xe2x80x9d accompanying his pointing motion to that direction. In order to respond to this question for value of truth or falsehood, xe2x80x9cthis directionxe2x80x9d must be recognized at S11 in FIG. 1. In order to recognize xe2x80x9cthis directionxe2x80x9d, a special sensor using technique of image recognition is necessary. As another example, in case the question for value of truth or falsehood as xe2x80x9cDo I push this button?xe2x80x9d is input to some help system, and the same problem accompanying this question occurs.
As mentioned-above, in natural language dialogue environment of the prior art, in order to reply to the question for value of truth or falsehood, status in which the apparatus is located is necessary to be recognized. Therefore, in addition to recognition of the input natural language, a special device, such as an expensive sensor to recognize the present status, is necessary. As a result, components become complicated and costs increase. Alternatively, even if the sensor is prepared, actual recognition of the status is sometimes impossible according to necessary information.
Furthermore, even if a solution for the question is obtained, in order to reply to the question for value of truth or falsehood, the system must compare the solution with the present status and calculate the value of truth or falsehood according to the comparison result. Therefore, calculation processing load of the system is high.
On the other hand, in case of the natural language dialogue apparatus for not accepting the question for value of truth or falsehood, the user can not speak the question for value of truth or falsehood for a proposition to be confirmed. Therefore, the user interface quality greatly goes down.
It is an object of the present invention to provide a natural language dialogue apparatus and a method to easily reply to the question for value of truth or falsehood without the recognition of present status and the decision of the value of truth or falsehood.
According to the present invention, there is provided a natural language dialogue apparatus, comprising: input analysis means for analyzing an input sentence of natural language from a user; blank data memory means for storing blank data to blank a question element being a decision object for a value of truth or falsehood in a question to ask the value of truth or falsehood; question data conversion means for deciding whether the input sentence is the question to ask the value of truth or falsehood, for estimating the question element in the input sentence if the input sentence is the question, and for converting the input sentence to question data in which the question element is blanked by referring to said blank data memory means; and problem solution means for retrieving a solution for the question element blanked in the question data, and for generating a response in which an answer for the value of truth or falsehood is not included according to the solution.
Further in accordance with the present invention, there is also provided a method for dialoging by natural language, comprising the steps of: analyzing an input sentence of natural language from a user; storing blank data to blank a question element being a decision object for a value of truth or falsehood in a question to ask the value of truth or falsehood; deciding whether the input sentence is the question to ask the value of truth or falsehood; estimating the question element in the input sentence if the input sentence is the question; converting the input sentence to question data in which the question element is blanked by referring to the blank data; retrieving a solution for the question element blanked in the question data; and generating a response in which an answer for the value of truth or falsehood is not included according to the solution.
Further in accordance with the present invention, there is also provided a computer readable memory containing computer readable instructions, comprising: instruction means for causing a computer to analyze an input sentence of natural language from a user; instruction means for causing a computer to store blank data to blank a question element being a decision object for a value of truth or falsehood in a question to ask the value of truth or falsehood; instruction means for causing a computer to decide whether the input sentence is the question to ask the value of truth or falsehood; instruction means for causing a computer to estimate the question element in the input sentence if the input sentence is the question; instruction means for causing a computer to convert the input sentence to question data in which the question element is blanked by referring to the blank data; instruction means for causing a computer to retrieve a solution for the question element blanked in the question data; and instruction means for causing a computer to generate a response in which an answer for the value of truth or falsehood is not included according to the solution.